Document summarizer

This application is a continuation of International Application No. PCT/US2023/020647, filed on May 2, 2023, which claims priority from U.S. Provisional Patent Application No. 63/337,323, filed on May 2, 2022. The disclosures of the above-referenced applications are expressly incorporated herein by reference in their entireties.

The disclosed technology relates generally to natural language processing from an automated computer-based system. More specifically, the disclosed technology includes the use of natural language processing techniques to automatically analyze the semantic context of an input text document and generate a summary of the input text document. While prior systems are able to generate document summaries, these systems lack the ability to account for various characteristics of the input text document in generating a summary. For example, prior systems do not rely upon visual characteristics or formatting structure of the input text document (e.g., multiple columns of text; page breaks; images interrupting the middle of a paragraph of text; etc.). As a result, prior systems often generate disjointed or incomplete summaries in the presence of such visual characteristics or formatting structure.

The disclosed embodiments are aimed at addressing the deficiencies of prior systems by generating summaries using a fusion of semantic context analysis relative to a particular input text document together with analysis of visual characteristics/formatting structure of the input text document.

The presently disclosed embodiments may include a computer readable medium including instructions that when executed by one or more processing devices cause the one or more processing devices to perform a method. The method may include receiving an identification of at least one source text document, loading text of the at least one source text document, and segmenting the text of the at least one source text document into two or more segments, wherein the segmentation is based on both formatting of the at least one source text document and semantic context of the text of the at least one source text document. The method may further include analyzing the segmented text of the at least one source text document, generating, based on the analysis, at least one summary snippet associated with one or more portions of the text of the at least one source text document, wherein the at least one summary snippet conveys a meaning associated with the one or more portions of the text, but includes one or more textual difference relative to the one or more portions of the text of the at least one source text document, and causing the at least one summary snippet to be shown on a display.

The presently disclosed embodiments may include a computer readable medium including instructions that when executed by one or more processing devices cause the one or more processing devices to perform a method. This method may include receiving an identification of at least one source audio or video with audio file, generating a textual transcript based on an audio component associated with the at least one source audio or video with audio file, and editing the textual transcript to provide a formatted textual transcript. The method may further include segmenting the formatted textual transcript into two or more segments, generating, based on analysis of the two or more segments, at least one summary snippet associated with the two or more segments, wherein the at least one summary snippet conveys a meaning associated with at least one of the two or more segments, but includes one or more textual differences relative to at least one of the two or more segments, and causing the at least one summary snippet to be shown on a display together with a representation of the at least one source audio or video with audio file.

The disclosed embodiments relate to a reading assistant system designed to generate summaries of input text documents. For many, the task of reading lengthy text documents can be arduous and time-consuming. The speed of reading can be slow due to the presence of learning difficulties such as dyslexia and, as a result, it can be particularly taxing to consume text documents. In other cases, the volume of textual material a particular user may wish to consume may exceed the amount of material the user can read within applicable time constraints. To address these issues, the disclosed embodiments automatically generate document summaries based on provided input text documents. One aim of the disclosed systems is to reduce the amount of time needed for a user to consume information included in textual documents.

Document Summarizer

is a schematic diagram of an exemplary system environment in which the disclosed reading assistant may be employed. For example, systemmay include a plurality of client devicesoperated by users. Systemmay also include a network, server, internet resources, cloud services, and databases. The components and arrangement of the components included in systemmay vary. Thus, systemmay include any number or any combination of the system environment components shown or may include other components or devices that perform or assist in the performance of the system or method consistent with the disclosed embodiments. Additionally, the disclosed reading assistant system may be implemented on any single component shown (e.g., a single mobile device or single PC included in client devices) or may be implemented in a network architecture (e.g., one or more features of the disclosed reading assistant systems and methods being implemented on a server, associated with one or more cloud services, etc. and having connectivity established with one or more client devicesvia network(e.g., a WAN, LAN, Internet connection, etc.)).

As shown in, client devicesmay include a variety of different types of devices, such as personal computers, mobile devices like smartphones and tablets, client terminals, supercomputers, etc. Client devicesmay be connected to a network such as network. In some cases, a usermay access the reading assistant and its associated functionality via the client devicewhich can display the user interface of the reading assistant.

represents an example operation flow associated with a reading assistant tool according to exemplary disclosed embodiments. For example, stepincludes acquiring text on which the reading assistant tool is to operate. The text may be acquired from various types of text files loaded or identified through an interface of the reading assistant tool.

Next, at step, the reading assistant tool can analyze and enrich the acquired text. For example, using AI-based models, trained neural networks, etc., the reading assistant tool can analyze the acquired text to do any of the following actions: identify and/or recognize entities described in the acquired text (even those identified by pronouns); summarize facts, information, argument, points, etc. associated with the acquired text; draw on external knowledge sources (e.g., databases, documents, etc. available via the Internet or other network) to augment information etc. conveyed by the acquired text; identify relationships between various types of entities associated with the acquired text; identify and/or extract keywords and key concepts from the acquired text; among other suitable tasks.

Based on the results of the reading assistant tool's analysis in step, the reading assistant tool can generate various types of outputs at stepto assist a user in working with/understanding the acquired text. For example, the reading assistant tool can generate summary snippets based on segments of the acquired text. The summary segments may convey key information or points associated with segments of the acquired text, while including one or more modifications to those segments. The modifications may include changing words, omitting words, substituting words, simplifying language complexity, removing phrases, adding words or phrases, etc.

In some cases, the reading assistant tool may generate an entities and relations graph, which graphically (or textually in some cases) identifies entities referenced in the acquired text and represents relationships between those entities. Information relating to the graphed relationships may be derived from the acquired text or may be augmented based on access to external knowledge sources (e.g., Internet databases, documents, etc.).

Stepmay include a semantic search capability and/or query-oriented summaries. For example, a user can enter search text into an input field (e.g., a query box, etc.), and the reading assistant tool can find words and phrases in a single source document or in multiple source documents provided as input that correlate with the contextual meaning of the input search text. The search text provide by the user can also be used for other purposes. For example, in some cases, the reading assistant/document summarizer tool can use the input search text as a guide in generating or updating one or more summary elements to emphasize certain semantic meanings, entities, relationships, facts, arguments, etc. indicated by the search text as of particular interest to a user.

As noted, in some cases, the user input received via a semantic search window may be used to analyze a collection of multiple source documents received or identified as input. For example, given a collection of documents and a user input (e.g., input text representative of semantic search query), the system can generate a summary of the information found in the collection of documents that is relevant to the user input. The user input may be provided as free text, and may include, among other things: a mention of a specific entity, a statement, a question, etc. One or more summaries generated based on the user input and the collection of documents may be linked to the source text/information included in the collection of documents, so that the user can jump from any portion of the summary to the relevant parts of a particular document or group of documents from which a summary snippet sentence was derived.

As part of the generation of summary snippets based on segments of the acquired text, the disclosed systems may rely upon determined spread scores associated with the snippets. For example, the document summary/reading assistant system may include one or more algorithms that compare a set of potential summaries to a particular text. The potential summaries may be ranked according to the degree to which the information contained in each summary is spread throughout the text (e.g., how frequently information from a summary appears in the text, how much of the text is related to or implicated by a summary, etc.). The higher the “spread score” for a particular summary, the more of the text's information is conveyed by the summary.

The determined spread scores for potential summaries can be used in determining which summaries to show to a user. For example, based on the spread scores for a set of potential summaries, the document summarizer/reading assistant system can rank the potential summaries. Those with higher rankings (meaning that more of the information in the source text is represented by those summaries as compared to other summaries with lower rankings) may be selected for showing to the user. For example, at step, the reading assistant tool may use the spread score and/or spread score rankings for a set of potential summary snippets to determine which of the potential summary snippets are shown to or made available to the user. In other words, when determining which summary snippet(s) to make available to a user, among multiple alternative summary snippets, the system may rely upon the spread score information to determine which snippet option(s) represent more of the information of a portion (e.g., a paragraph, section, page, etc.) of the input source text.

At step, the reading assistant tool may also offer content-based completion functionality. For example, via an interface associated with the reading assistant tool, the system may offer text suggestions, as the user inputs text (e.g., capturing notes or thoughts of the user relative to the input text and/or the generated summary snippets). In some cases, the user may augment the generated summary snippets by inputting additional text into one or more summary snippets. As the user enters text, the system may offer suggestions for content completion. Text suggestions offered to the user may include single words or short phrases. In other cases, however, the system may offer text suggestions in the form of one or more complete sentences. These text suggestions can also be based on the context and content of source text from one or more input text documents loaded into or identified to the reading assistant tool (or based on externally accessible sources).

At step, the reading assistant tool may also offer side-by-side read and write capability. For example, any of the summary elements generated based on the text analysis performed in stepmay be shown in an interface of the reading assistant tool in a side-by-side relation to source text to which the summary elements relate. The interface of the reading assistant tool may also provide a text editor window such that the user can draft text while having proximate access to the source text and summary elements relating to the source text.

Returning to step, an interface of the reading assistant tool may include any suitable interface for loading or identifying text documents. For example, activation of the reading assistant tool may cause a window, such as windowshown into be shown on a display. Windowmay include an active areato facilitate identification of source text documents to the reading assistant tool. For example, a user may drag and drop text files into active areato load documents into the reading assistant tool. Alternatively or additionally, a user may click on a “browse” link to access a file system associated with one or more storage devices and may select one or more text files from the file system for loading into the reading assistant tool. Further, a user may type or copy and paste an address, such as a URL, into address fieldin order to identify to the reading assistant tool one or more documents to access and load. Any of these techniques can be used alone or in combination to load documents into the reading assistant tool, especially as the reading assistant tool can load and operate upon multiple documents from multiple different sources or storage locations in a single session.

Upon loading one or more text documents, the reading assistant tool can analyze the loaded text documents (step) and can generate one or more summaries relative to the loaded text documents. The generated summaries can be shown to the user in any suitable format.provides a block diagram representation of a generic summary windowthat may be included in an interface associated with the disclosed reading assistant tool. Windowmay be arranged with various different layouts and may include various combination types and combinations of display windows, scroll bars, summary snippet bubbles, text entry fields, virtual buttons, toolbars, drop down menus, etc. In the particular example shown in, interface windowincludes an analysis panel, a text review panel, a summary panel, and a writing panel.

Each panel type, along with its exemplary associated functions and features, is discussed in more detail below. In general, however, analysis panelmay provide one or more portals to results of analysis performed by the reading assistant tool in step. Such results may include: information relating to identified entities and entity relationships; compressed text summaries; information extracted from external knowledge sources; keyword and concept extraction; among others.

Text review panelmay include a reproduction of at least a portion of the text analyzed in one or more input/source text documents loaded into the reading assistant tool. Text shown in the text review panelmay include highlighting, underlining, bolding, or other types of emphases to indicate what portions contributed to summaries, such as summary snippetsincluded in summary panel. Writing panelcan receive text entered by a user, text copy and pasted (or drag and dropped) from text review panelor from text snippets, for example.

Interface windowmay include various other types of information or functionality. For example, interface windowmay identify a document's meta-datum (e.g., a document title) to identify the file name or other document identifier associated with the particular source text file (or a project text file including text from multiple source text files) under review.

provides an example of a summary interface windowthat may be generated by the reading assistant tool. In this example, windowincludes a text review panelthat includes a reproduction of a portion of a source text document (i.e., and article entitled, “Seven Legal Questions for Data Scientists”) loaded into the reading assistant tool. The name of the source text document is also shown in title field.

After analyzing the source text document and generating one or more summaries relative to the document, the reading assistant tool can show the generated summaries on a display. In the example of, a number of summaries fieldindicates how many summaries the reading assistant tool generated during the analysis phase, and the summaries can be shown in a summary window. In this example, the summaries are shown in summary snippet boxes, however, any other suitable format (e.g., text bubbles, bulleted outline, etc.) may be used to show the generated summaries on a display.

Each summary generated may be based upon at least some portion of the text in a source text document loaded into the reading assistant tool. In the example of, the reading assistant tool may be equipped to identify to the user a portion or portions of the source text document(s) that contributed to the generation of a particular summary. For example, as shown in, text relating to a particular summary can be highlighted, underlined, bolded, etc. to indicate that it relates to at least one generated summary. A highlights toggle barmay be provided to enable the user to toggle on and off the highlighting of text used in generating one or more summaries.

Links between generated summaries and the associated text based on which they were generated may be indicated in any suitable manner. For example, as shown in, a generated summary, including a summary snippetshown in summary snippet box, may be displayed next to its corresponding text in a source document (e.g., the text based on which the summary snippet was generated). In this example, the reproduced text from the source text document is shown in text review panelwith highlighted text. The proximity of summary snippetto highlighted textcan indicate to a user that highlighted textcontributed to the generation of summary snippet. In some cases, especially where the density of generated summary snippets is higher, other indicators, such as lead lines, color coding, etc. may be used to indicate relationships between generated summaries and text used to generate the summaries.

Interface windowmay include various tools and controls to assist a user in efficiently reviewing and understanding content included in the source text documents loaded into the reading assistant tool. For example, as indicated by the number of summaries field, in the example of, the reading assistant tool has generatedsummaries based on its analysis of at least the loaded source text document partially reproduced in text review panel. To review the generated summaries, the user can interact with a scroll bar (not shown). For example, dragging a scroll bar downward may cause the text shown in text review panel, as well as the generated summaries shown in summary review panelto migrate upwards on the display screen to bring additional text from source document and additional generated summaries into view on the display. In this way, a user can quickly scroll through the generated summaries and develop a good understanding of the source document through review of the generated summaries alone. Should the user wish to clarify any details or to gain further context relative to any particular generated summary, the side-by-side display of source text (optionally with highlighting) and corresponding summaries may enable the user to quickly access the text in the source document most pertinent to a particular summary. And, if the user wishes to review the complete text of the source document, it is available and shown in the text review panel. To further illustrate the original text to which a generated summary relates, the reading assistant tool may include a highlight baridentifying a portion of the original text for which one or more summaries were generated.

As noted above, a component of the analysis performed by the reading assistant tool in stepis the identification of entities referenced by source text documents and the determination of relationships among those entities as conveyed by the source text documents (and optionally as augmented by external knowledge sources). Through analysis of the source text documents, for example, the reading assistant tool can automatically create a knowledge graph of entities (e.g. a person, organization, event, process, task, etc.) mentioned/referenced in unstructured text in source text documents. The knowledge graph may include, among other things, entities, relations between entities, information about the entities, and instances of each entity in the text. The different instances of each entity are extracted and associated with the entity even if the entity was diversely and implicitly referenced (including reference by a pronoun, semantic frames where the entity has a semantic role not explicitly stated, etc.). The knowledge graph can also be generated or augmented based on access to external knowledge sources (e.g., accessible Internet sources, private knowledge bases, or knowledge bases local to the reading assistant tool). Using such sources can provide further information on the entities and the relations among the entities.

In some cases, the knowledge graph refers to the entity relationships identified and maintained internal to the models/networks associated with the reading assistant tool. In other cases, the knowledge graph may be provided to a user. For example, a user may click on a knowledge graph portal (such as the “Entities and Relationships” active region/clickable area/button shown in), and the reading assistant tool may show on the display the results of its entity and relationships analysis relative to the source text documents. In some cases, the knowledge graph may be represented to a user in a graphical format (e.g., entities identified in boxes or bubbles that may be arranged, connected by lines, associated with symbols, etc. to convey information about relationships, hierarchies, etc. among the identified entities). In other cases, the knowledge graph may be represented to the user in a text-based format (e.g., list, outline, etc.).

Other features or functionality of the reading assistant tool can also enable the user to interact with loaded source text documents, especially with respect to entities identified or referenced in the source text documents. For example, in some embodiments, the user can select a span of text in a loaded source text document, and in response, the reading assistant can display to the user the entities referenced in the selected span of text. In another example, a user may mark/identify multiple documents (e.g., by clicking on, highlighting, etc. icons or filenames representative of the documents), and in response, the disclosed reading assistant/document summarizer system may generate an entity graph indicative of entities identified across the multiple documents. The entity graph may include a comprehensive list of entities referenced in the multiple documents and may indicate which of the documents refer to which entities. The entity graph may also include information conveying how many times each identified entity is referenced by each of the multiple documents.

Additionally or alternatively, the reading assistant tool can enable the user to view or navigate to other instances of the same entity or to other related entities in the source text documents. Further, the reading assistant tool can enable the user to view information about the entity that the tool extracted from the source text documents or acquired from external sources.

provides another example of a summary window interfaceprovided by an embodiment of the described reading assistant tool. Summary window interfaceincludes a text review panelshown in side-by-side relationship to a summary review panel. In this example, three summaries, including summary snippets, have been generated based on text from the source document currently shown in the text review panel. As an additional feature, a highlight barmay be configured to identify (e.g., using color coding, line thickness, etc.) portions of the source text for which the reading assistant tool has generated at least one summary.

In some cases, as described above, the reading assistant tool can automatically generate one or more summaries based on loaded source text without additional input from a user. In other cases, however, the reading assistant tool may provide a guided summarization feature with which the user may guide the summaries generated by the reading assistant tool through supplemental input provided to the reading assistant tool. For example, after (or in some cases before) the reading assistant tool automatically generates one or more summaries based on loaded source text, a user may provide supplemental text input to the reading assistant tool (e.g., via a text input window). The reading assistant tool can update generated text summaries (or generate new text summaries) based on the text input provided by the user.

The text input provided by a user can be free text input. The text input, for example, can specify a subject or theme of interest; identify, indicate, or reference, among other things: entities (e.g a particular person, organization, event, process, task), entity types (e.g. ‘organizations’, ‘managers’, ‘meetings’, ‘requests’), topics (e.g. ‘finance’, ‘sales’, ‘people’), or concepts (e.g. ‘positive,’ ‘good,’ ‘happy,’ etc.). In response to receiving the free text input from the user, the reading assistant tool can generate one or more summaries based on the loaded source text as well as the text input received from the user. The reading assistant tool can further highlight instances in one or more loaded source documents related to the free text entered by the user. The reading assistant tool can also select information from the loaded source text that pertains to the subject or theme, etc., of the user's text input even if none of the input text, or its morphological modifications, is found in verbatim in the text spans containing the information. The reading assistant tool can then include the selected information into one or more generated summaries, and the summaries can be organized based on the subject, theme, etc. conveyed by the user's input text.

provides a block diagram representation of the process flow of the guided summarization feature of some embodiments of the disclosed reading assistant tool. At step, the reading assistant tool receives text input from the user. At step, the reading assistant tool analyzes the loaded source text documents and identifies sections of the source text relevant to the subject, theme, concept, etc. implicated by the user's text input. At step, the reading assistant tool generates one or more summaries based on both the user's text input and the text of the source text documents. At step, the reading assistant tool shows to the user (e.g., through an interface window on a display) the locations in the source text documents of text sections relevant to the user's input. The reading assistant tool also shows to the user the summaries generated based on the source text and the user's text input.

illustrates an example of the guided summarization functionality of embodiments of the disclosed reading assistant tool. For example, interface windowshows an output of the reading assistant tool before receiving text input from the user, and interface windowshows an output of the reading assistant tool after receiving text input from the user. Specifically, as shown in the example of, the interface of the reading assistant tool may include a user text entry field. As shown in interface window, user text entry fieldis blank and only includes the reading assistant tool's prompt, “Summarize according to . . . ”. With no user text input to guide the summarization function, the reading assistant tool analyzes the loaded source text documents and generates summaries. In this case, two summary snippets are shown, and scroll barshows a current location relative to the source text document and locations of all summaries generated relative to the source text document. The two currently displayed summaries, generated without text input from the user, read:

“In qualifying plans with high deductibles, individuals can contribute pre-tax money to a Health Savings Account. As deductibles rise, more plans are becoming eligible for HSAs.”

“Unspent money can be invested in the account and earn interest. HSA deposits are estimated to reach $75B in 2020.”

Interface windowrepresents how the reading assistant tool can rely upon user text input to guide the summaries generated relative to the source text document. For example, as shown in user text entry window′, the user has entered the phrase, “Health expenses.” In response, and based on the user's text input, the reading assistant tool generates new summaries (e.g., updated summaries) relative to the source document text. For example, relative to the same section of the source text document shown in both windowsand, the reading assistant tool, in response to receiving the user text input, has generated a new summary. Not only is there one less summary relative to the same text passage, but the summarydiffers from the summaries. Specifically, summaryreads:

“Health Savings Accounts allow contributing pre-tax money to a health expenses account.”

Notably, the newly generated summaryconveys a meaning similar to a portion of the first of summaries, but summarymore prominently features the subject “health expenses” of the user's entered text. In addition, the reading assistant tool has linked the concept “health expenses” with “HSAs” and has referred to HSAs as “health expenses accounts” rather than “health savings accounts,” to which the HSA acronym refers. Of course, a primary use for an HSA is to cover health expenses, which is the relationship gleaned by the reading assistant tool based on its training and/or its analysis of the source text documents. This connection provides one example of the reading assistant tool's capability for linking subjects, entities, concepts, etc. even where there is not a literal textual link for the connection.

As shown in, the reading assistant tool can also identify to the user the locations of summaries, relative to the source document text, that are particularly relevant to the user's entered text. For example, in the example represented in, the reading assistant tool has added in interface windowhighlighted tick marksto indicate where, relative to the source text, the reading assistant tool generated summaries relevant to the user's entered text, “Health expenses.” And, as shown in window′, the current location of scroll bar′ is shown as overlapping with one highlighted summary (i.e., the location relative to the source text of generated summary).

illustrates an example of another feature of some embodiments of the reading assistant tool. Specifically, in some cases, the reading assistant tool may be equipped with the ability to assist the user in drafting text by analyzing user-entered text and then suggesting supplements to the entered text, providing text re-write suggestions, etc. As the basis for the supplement and/or re-write suggestions, the reading assistant tool can draw upon facts, information, concepts, etc. referenced in one or more source text documents loaded into the reading assistant tool. The reading assistant tool can also draw upon facts, information, concepts, etc. referenced in one or more external databases as the basis for the supplement and/or re-write suggestions.

The reading assistant tool offers an integrated flow for composing a written document while a user interacts with the reading assistant. For example, as shown in, the reading assistant tool may include an interface window, which includes a source text review panel, a summary review panel, and a text composition panel. As an aside, the panels of interface windowmay all be re-sized by the user depending on which section the user is most interested, in which section the user is currently working, etc. Text review paneland summary review panelcan operate similarly to text review panel and summary review panel described relative to. For example, based on analysis of the loaded source text document, represented in text review panel, the reading assistant tool can generate one or more summaries, such as summary snippet, based on the source text and based on any entered user input text (optionally entered via user text entry field).

In the example of, text composition windowmay be used by the user as a text editor to draft document text. In some cases, the user can copy and paste into text composition windowtext obtained from text review paneland/or from summary review panel. In addition, the user can also introduce free text edits into text composition window. As the user enters free text, the reading assistant tool can analyze the user's entered text and, similar to the functionality of the reading assistant tool described herein, can provide suggestions to the user for re-writing portions of user-entered text or for supplementing the user-entered text. The reading assistant tool's suggestions are based on both the text entered by the user and also on the loaded document source text and/or summary text generated by the reading assistant tool.

represents an example of this functionality. Specifically, in text composition window, the user has entered text. Textmay include sections copy and pasted from text review paneland/or from summary review panel. Textmay also include free text entered by the user. In this example, as the user was composing the last sentence shown in text passage, the reading assistant tool offered suggestionfor completing the sentence. That is, the user had entered the phrase, “The percentage of workers with HSAs increased,” and in response, the reading assistant tool suggested the phrase, “by 280% in the past decade” to complete the sentence. The reading assistant's suggestion was based on concepts conveyed in both the user's entered text and in the source document text or summary text. For example, entry of the phrase “The percentage of workers with HSAs increased” prompted the reading assistant tool to refer to the facts, entities, relationships, etc. established for the source text document based on the analysis of that document to determine if the source document or relevant summaries contained any information relating to the percentage of workers with HSAs. Based on the user's entered text and its prior analysis of the source text document and generation of corresponding summaries, the reading assistant tool identified the fact that 23% of workers in 2019 had an HSA, compared to just 6% in 2009, which equates to a 280% increase. Thus, the reading assistant's suggestion for completing the user's sentence was drawn from facts and context conveyed by the user's text, as well as facts and context associated with the source document text/relevant summary. Notably, however, the text suggestion offered by the reading assistant tool was not framed in terms of the underlying percentages of workers with HSAs data, as included in the source text/summary. Rather, because the user's text referenced an “increase,” the reading assistant tool was able to link the concept of an “increase” to an increase amount (i.e., 280%) in the underlying percentages between 2009 and 2019. The reading assistant tool was also able to link a difference in years (i.e., 2009 to 2019) to the concept of a “decade” and a comparison of a current time (e.g., 2020) to the years identified in the source text/summary to determine that 2009 to 2019 represents the decade before the current year. In view of these links and determined relationships, the reading assistant tool expressed the suggested sentence ending not in the literal facts/text appearing in the source text/summary, but rather in terms of a more complex concept, “in the past decade,” which accurately, but differently, conveys the meaning/information included in the source text/summary.

To assist the user, the reading assistant tool can identify the source text or summary text serving as the basis for suggested re-write options or suggested text supplements. In the example of, suggestionincludes highlighting to identify the generated suggestion to the user. The reading assistant tool can also display the text from one or more summary snippets, such as snippet(or text from the source document) on which the suggestion was based. In the example of, suggestionis shown in proximity to snippet(and optionally associated text from the source document) to identify to the user the information used as the basis for suggesting the phrase, “by 280% in the past decade.”

The reading assistant tool can also offer the user the option to select a boxto automatically link the text suggestion to the source text or texts from which it was derived (an auto-citation function). The text suggestions offered by the reading assistant tool may include facts, direct quotes, paraphrased information, summarized information, etc. derived from the loaded source text documents and/or derived from externally one or more accessible documents or knowledge bases (e.g., via the Internet). The reading assistant's text completion and generation suggestions can also be modulated according to a currently active page of the source document, based on currently active summaries (e.g., those source document pages and summaries currently shown in an interface window associated with the reading assistant tool), or based on current text selections from the source document made by the user.